Audit

Skill sevenbelowllc/the-library/skills/audit

Three-way gap analysis: canonical specs vs vault wiki vs actual code. Finds missing features, claimed-but-broken, and undocumented implementations.From its SKILL.md

Install

npx -y skills add sevenbelowllc/the-library --skill audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.8 KB, as published. Nobody here has run it

library:audit — Spec vs Code Gap Analysis

The skill that prevents "features claimed complete but actually broken."

When to Use

Before starting a new development cycle
After claiming a milestone is complete
Periodically to detect drift between spec, docs, and code
When onboarding to understand true project state

Process

Step 1: Load Sources

Specs: Call library_config_get for reading_room.path, then read the markdown files in <reading_room.path>/specs/
Vault: Call library_vault_parse for wiki articles and tags
Code: Read project structure, test files, key source files

Step 2: Extract Claims

From specs: what SHOULD exist (features, endpoints, behaviors) From vault wiki: what is CLAIMED to exist From code: what ACTUALLY exists (files, functions, tests, routes)

Step 3: Three-Way Diff

Spec Says	Vault Says	Code Shows	Verdict
Feature X	Implemented	Code exists, tests pass	VERIFIED
Feature X	Implemented	Code exists, no tests	UNVERIFIED
Feature X	Implemented	Code missing	CLAIMED-BUT-BROKEN
Feature X	Not mentioned	Code exists	UNDOCUMENTED
Feature X	Not mentioned	Code missing	MISSING
Not in spec	Implemented	Code exists	SCOPE CREEP

Step 4: Dependency Analysis (if Graphify enabled)

For each broken/missing item:

Call library_graph_query: "What depends on <item>?"
Report cascade impact

Step 5: Generate Report

Output structured gap report:

VERIFIED: N items (green)
UNVERIFIED: N items (yellow — need tests)
CLAIMED-BUT-BROKEN: N items (red — urgent)
MISSING: N items (red — not started)
UNDOCUMENTED: N items (yellow — need vault update)
SCOPE CREEP: N items (orange — discuss)

Step 6: Create Tasks (optional)

Ask: "Create PM tasks for gaps?" If yes, call library_pm_create_task for each actionable gap.

Subagent Orchestration

For large codebases, delegate file-heavy inventory work to parallel subagents:

Step 2 (Inventory) — Dispatch 3 parallel Explore subagents:

Agent 1 (Specs): "Inventory the specs/ directory. List each file, extract key claims and requirements. Return a structured summary under 500 words."
Agent 2 (Code): "Inventory the codebase. List GraphQL modules, services, migrations, and key patterns. Return a structured summary under 500 words."
Agent 3 (Vault/Wiki): "Inventory the vault wiki articles. List each article, extract [VERIFY], [CONFLICT], [PLANNED] tags. Return a structured summary under 500 words."

Main context receives ~1500 tokens of structured summaries instead of ~15000+ tokens of raw file reads.

Step 5 (Verification) — Dispatch targeted verification agents:

For each load-bearing claim in the gap analysis, dispatch a focused Explore subagent:

"Does compliance-core hardcode DRAFT status anywhere?"
"Which migrations create the agent_jobs table?"

Fallback

If the Agent tool is not available (e.g., non-Claude Code environments), fall back to sequential file reads in main context. The skill works either way — subagents are an optimization, not a requirement.

Token Budget

Weight: Medium (was Heavy before subagent refactor) Estimated context cost: ~2000 tokens with subagents, ~15000+ without Subagent delegation: Yes — inventory and verification steps

MCP Tools Used

library_vault_parse — vault state
library_graph_query — dependency analysis
library_pm_create_task — create tasks for gaps
library_config_get — reading room path (specs live in <reading_room.path>/specs/), project keys